This should be fairly quick because the slowest parts are the setup (copying files, fetching files, bootstrap, etc). Helps avoid regressions like:
Split out of !956 (merged) so we can commit it once the regression is merged into the gstreamer monorepo.